Suchir Balaji worked at OpenAI from 2021 until August 2024, helping gather and organize internet data used to train GPT-4 and contributing to the precursor WebGPT. In an essay published in October 2024 and a New York Times interview on October 23, he argued that systems such as ChatGPT were trained on competitors’ copyrighted work and generated outputs capable of imitating or substituting for the original products. His analysis applied the four-factor fair-use test and concluded that generative AI could undermine the commercial viability of the people and organizations that produced its training data.
The allegations came as authors and publishers, including The New York Times, were suing OpenAI for copyright infringement. On November 18, 2024, Times attorneys identified Balaji as a potential source of relevant documents, and he said he would testify, although reporting noted that he had not disclosed previously unknown internal information. OpenAI maintained that its models were grounded in fair use and trained on publicly available data. Stanford professor Mark Lemley disagreed with Balaji’s legal conclusion, while attorney Bradley Hulbert said legislation might ultimately be needed. Balaji died on November 26, 2024; after public speculation, the San Francisco police and medical examiner concluded on February 14, 2025, that he died by suicide and found no evidence of foul play.
